Research in large, complex networks is becoming increasingly prevalent in many fields, such as neuroscience, social sciences, and economics. In many cases, researchers want to identify the topology of the network, such as which cells are connected to which other cells in the brain or who is influencing whom in a large social network.
In this talk, I will present a general framework for identifying and succinctly representing the topology of networks of causally interacting processes. It is grounded in information theory and can be applied to any network of causally interacting stochastic processes. The framework consists of three components. The first involves well defined graphical models to represent the network. The second entails efficient algorithms to identify or approximate the structure. The third describes a provably-good estimation procedure to compute necessary statistical quantities. We demonstrate the effectiveness of our methods through simulations and analyzing both neuroscience and Twitter data.
